论文部分内容阅读
Citrus is one of the major agricultural products in Florida, which contributes to approximately 65% of the overall citrus production in the United States.Citrus industry plays a vital role in Floridas agriculture domain.Accurate and efficient recognition and determination of green immature citrus fruit at an early stage under natural complex illumination would help the growers to plan application of nutrients during the fruit maturing stages and estimate their profit prior to harvesting period.This also allows citrus growers to plan in advance so they can determine how much labor is needed during the harvesting period and well allocate labor depending on the yield prediction.The growers can have an estimation of their profits for the whole grove prior to harvesting period, and this would be one of the most important factors guiding the price trend of the market.The precise recognition of the fruit in field is the precondition of yield production and fruit picking for harvesting robots.So developing a robust algorithm to recognize fruits from its complicated natural growing environment effectively is the initial demand in the precision agriculture.However, occlusion, varying illumination, and color similarity with the background make green citrus fruit identification a very challenging task.The goal of this research was to develop a robust and fast algorithm to detect and count immature green citrus fruit in individual trees from color images acquired under natural outdoor conditions.A total of 120 images were obtained from an experimental citrus grove in the University of Florida, Gainesville, Florida, USA.In order to make the fruit recognition system much more efficient to mect the requirement of on-line detection while avoiding image distortion, a nearest-neighbor interpolation method was used to resize images from the resolution of 3648×2736 pixels to 912 × 684 pixels.Fifty-four images were used for a training set, and the remaining sixty-six images used for validation.In order to overcome recognition challenges in uncontrolled environments caused by uneven illumination conditions, similar background features, and partial occlusion, multiple features such as color,shape and texture features of the different objects (immature green citrus, leaves, soil, sky, twigs and branches) were used for feature extraction to find differences between green citrus and other objects.Firstly, normalized chromatic color R-B in OHTA color space and equalized histogram threshold based method were combined to filter non-fruit pixels, while keeping as many as possible fruit pixels.Watershed segmentation was utilized to separate overlapped fruits then shape feature was used to detect potential fruits.It is inevitable that some non-fruit regions would be recognized as fruit because of the color similarity and occlusion.To remove falsely detected fruits, texture feature was carried out to distinguish fruit and non-fruit regions.After removal of false positives,multiple circles merged based on the distance between each centers.The proposed method can provide a more efficient way for green citrus identification in a grove under natural outdoor conditions.